Getting Started
For the automatic pipeline: Contact us to onboard.
For the SplitStep App: Contact us for beta access.
Technical Requirements
For optimal tracking accuracy, the camera needs to be positioned as follows:
- Centered behind the baseline: The camera should be positioned more or less centered at the back of the court, aligned with the center service line.
- Elevated position: The camera should be elevated above court level. Higher is better. The more elevation, the better the tracking accuracy.
- No side angles: We do not support side-angle footage. The camera must be positioned behind one of the baselines.
Important: Accuracy decreases when the camera is positioned off the center line. The further off-center the camera angle, the less accurate the tracking data will be.
All of them! Just get us the video. We run all our algorithms in the cloud, so we're not tied to any specific camera ecosystem.
Videos submitted through our API should meet the following guidelines:
- Resolution: 1080p (1920×1080) or higher
- Frame rate: 30 fps minimum, 60 fps recommended
- Format: MP4 (H.264) preferred
- File size: Less than 5 GB
- Length: Adjust StartTime and EndTime to cover complete games, consistent with the set game scores you provide
- Mode: Singles only — doubles matches are not supported at this time
Capabilities
Singles only — doubles matches are not supported at this time.
Pickleball and other racket sports are in the pipeline and will be worked on soon.
Either! SplitStep works great for both practice sessions and competitive matches.
Output & Integration
Automatic pipeline: We output JSON files with tracking and event data converted to real-world coordinates (and pixels). This includes player positions, strokes, and ball bounces, along with ball flight metrics such as speed, spin, and net clearance. Frame-by-frame tracking for players and ball is available on request, and we can tailor the output format to match your needs.
Score context (point, game, and set) is also included, currently in beta. See the API documentation for the full schema.
Absolutely! This is what the automatic pipeline is designed for. You hit our endpoint, and we send your results back ready to integrate into your workflow.
This is where the fun begins!
The data is designed to fit into your current tools, or allows you to build scalable models that no longer rely on manually tagged data. You can now build your shot quality model over hundreds of games!
Get in touch if you'd like to brainstorm. We're happy to help.
Performance
If you're using our automatic pipeline, it's typically less than 1 hour for a 2-hour match.
We're always striving to improve, and video quality can drastically affect results. With an ideal camera setup, we achieve approximately 95% accuracy for stroke detection.
Ball bounce detection accuracy metrics will be added soon.
Security
Yes. Your videos and your results are never shared with other customers, and we never sell or license them to third parties.
Videos you upload are processed and stored on our own encrypted cloud infrastructure, accessible only to us and to the service providers who help us run it under confidentiality obligations. We do use uploaded footage to improve our models. If you'd rather we didn't, that's something we can agree in a commercial contract — see our Data License Agreement.
See our EULA and Privacy Policy for full details.
About
We've been working in this space consulting for hockey, soccer, and NBA teams. We've learned a lot, including what you can do with raw tracking data.
But in tennis, getting that raw data has always been a challenge. SplitStep unlocks that first hurdle.
A split step is a small hop timed just before your opponent hits the ball. You land lightly on the balls of your feet in a balanced, athletic stance, ready to push off instantly in any direction.
It's essential in racket sports (tennis, pickleball, badminton) for optimal balance and faster reactions to the ball's trajectory.